Big data health analysis method and system in medical and health care fusion environment

Abstract

The invention relates to the technical field of data analysis, and discloses a big data health analysis method in a medical and health care fusion environment, which comprises the following steps: acquiring to-be-analyzed data in the medical and health care fusion environment, carrying out cross-modal fusion, and carrying out word vectorization processing on the fused data to obtain word vector data; performing data clustering on the word vector data, calculating the clustering precision of the clustered data, and performing clustering optimization on the clustered data to obtain precision clustered data; performing data mining on the precision clustered data to obtain supplementary data, performing data screening on the supplementary data to obtain screened data, and taking the precision clustered data and the screened data as target analysis data; and querying user health features and user information of the medical and nursing user corresponding to the target analysis data, constructing a user portrait of the medical and nursing user, performing health association analysis on the medical and nursing user, and constructing a health analysis report of the medical and nursing user by using a health association analysis result. According to the invention, the big data health analysis level under the medical-care fusion environment can be improved.

Description

Technical Field

[0001] The present invention relates to a big data health analysis method and system in a medical and nursing integrated environment, belonging to the technical field of data analysis. Background Art

[0002] In the context of medical and nursing integration, it involves a lot of data, such as basic health indicator data, disease diagnosis and treatment records, medication status, self-care ability assessment, psychological state monitoring and other large amounts of data. Building a big data health analysis platform for medical and nursing integration can help medical staff and elderly care service personnel to comprehensively and systematically collect and integrate these complex data, and then deeply analyze the health status and living conditions of medical and nursing users. Through such a platform, potential health risks can be warned in advance, so that medical staff can take corresponding intervention measures in time. At the same time, it can also help medical and nursing service personnel to adjust care plans more targeted, thereby improving the quality and accuracy of medical and nursing integration services, better guaranteeing the healthy life of medical and nursing users, and facilitating information exchange and collaborative work between staff in different professional fields such as medical care and elderly care, and improving the collaboration efficiency of the entire medical and nursing integration team.

[0003] At present, for big data health analysis in the medical and nursing integration environment, most of them use conventional data mining, analysis technology and some general information to build a system with certain data integration and simple analysis functions to try to achieve the goal of big data health analysis. However, there are many problems with such conventional practices. Because the data sources in the medical and nursing integration environment are extremely wide and the data types are rich and diverse, there are both structured data such as physical examination indicators and unstructured data such as descriptions of daily health feelings of medical and nursing users, and the data formats and standards of different data sources are often inconsistent. This leads to increased difficulty in pre-processing such as data cleaning and conversion when conducting big data analysis, and many key information may not be accurately extracted and effectively used. In the case of large and complex data volumes, the analysis results are prone to deviations, which makes it difficult to achieve ideal results in big data health analysis in the medical and nursing integration environment. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a big data health analysis method in a medical and nursing integration environment, which can improve the level of big data health analysis in the medical and nursing integration environment.

[0005] To achieve the above purpose, the present invention provides a big data health analysis method in a medical and nursing integrated environment, including:

[0006] Acquire the data to be analyzed in the medical and nursing integration environment, extract the cross-modal features of the data to be analyzed, perform cross-modal fusion on the data to be analyzed based on the cross-modal features to obtain fused data, construct a word vector space of the fused data, and use the word vector space to perform word vectorization processing on the fused data to obtain word vector data;

[0007] Constructing a small sample of the word vector data, calculating a local density extreme value of the small sample, clustering the word vector data based on the local density extreme value to obtain clustered data, calculating the clustering accuracy of the clustered data, and performing clustering optimization on the clustered data based on the clustering accuracy to obtain precision clustered data;

[0008] Performing data mining on the precision clustering data to obtain supplementary data, constructing a true discriminant network for the supplementary data, performing data authenticity evaluation on the supplementary data using the true discriminant network, performing data screening on the supplementary data using an evaluation result of the data authenticity evaluation to obtain screened data, and using the precision clustering data and the screened data as target analysis data;

[0009] Query the medical and nursing users corresponding to the target analysis data, extract the user health characteristics of the medical and nursing users in the target analysis data, query the user information of the medical and nursing users, use the user information and the health characteristics to build a user portrait of the medical and nursing users, use the user portrait to perform a health assessment on the medical and nursing users, use the assessment result of the health assessment and the user information to perform a health association analysis on the medical and nursing users, and use the analysis result of the health association analysis to build a health analysis report for the medical and nursing users.

[0010] Optionally, extracting cross-modal features of the data to be analyzed includes:

[0011] Performing data cleaning on the data to be analyzed to obtain cleaned data;

[0012] dividing the cleaned data into visual data, text data and numerical data;

[0013] Performing visual feature processing on the visual data to obtain visual features;

[0014] Performing text feature processing on the text data to obtain text features;

[0015] Performing statistical feature processing on the numerical data to obtain numerical features;

[0016] Determine the cross-modal features of the data to be analyzed according to the visual features, the text features, and the numerical features.

[0017] Optionally, the using the word vector space to perform word vectorization processing on the fused data to obtain word vector data includes:

[0018] Performing word segmentation processing on the fused data to obtain word segmentation data;

[0019] After the word segmentation data are numbered, a vocabulary table of the word segmentation data is constructed;

[0020] Using the vocabulary, constructing a segmentation vector of the segmentation data after segmentation numbering;

[0021] After the word segmentation vector is input into the word vector space, word vector data is obtained.

[0022] Optionally, constructing a small sample of the word vector data includes:

[0023] Randomly sampling the word vector data to obtain sampled word vector data;

[0024] Constructing an initial word vector sample of the word vector data;

[0025] The sample error of the initial word vector sample is calculated using the following formula:

[0026]

[0027] Among them, ME represents the sample error, A represents the number of sampled word vector data, B represents the number of initial word vector samples, and G (i,j) represents the probability that the i-th sampled word vector data is assigned to the j-th initial word vector sample,

[0028] According to the sample error, the initial word vector sample is updated to obtain a small sample.

[0029] Optionally, the calculating the local density extreme value of the small sample includes:

[0030] Query the original word vector corresponding to the word vector data corresponding to the small sample and the small sample word vector corresponding to the small sample;

[0031] Constructing the spatial coordinates of the small sample;

[0032] The vector distance between the original word vector and the small sample word vector is calculated based on the spatial coordinates using the following formula:

[0033]

[0034] X=(x2、y2、z2)

[0035] Y=(x1, y1, z1)

[0036] Among them, VD represents the vector distance, X represents the small sample vector, Y represents the original word vector, (x2, y2, z2) represents a point position of the small sample vector in the spatial coordinates, and (x1, y1, z1) represents a point position of the original word vector in the spatial coordinates.

[0037] The local density extreme value of the small sample is determined according to the vector distance.

[0038] Optionally, calculating the clustering accuracy of the clustered data includes:

[0039] The clustering accuracy of the clustered data is calculated using the following formula:

[0040]

[0041] Among them, NMI represents the clustering accuracy, T represents the number of clusters of clustered data, n represents the number of clusters generated by the clustering algorithm, and E i Indicates the total number of data points in the i-th cluster in the clustering algorithm result, E j Represents the total number of data points in the jth cluster in the clustered data, Represents the total number of data points that appear in both the i-th cluster of the algorithm result and the j-th class of the true result.

[0042] Optionally, constructing a true discriminant network for the supplementary data includes:

[0043] Designing a discriminative network architecture for the supplementary data;

[0044] Performing data labeling on the supplementary data to obtain labeled data;

[0045] Dividing the labeled data into training data, verification data, and test data;

[0046] Using the training data, the verification data and the test data, the discriminant network architecture is preliminarily trained to obtain a trained discriminant network;

[0047] The loss value of the training discriminant network is calculated using the following formula:

[0048]

[0049] Among them, MSE represents the loss value, M represents the number of training data, and H i represents the true label of the training sample corresponding to the training network, Represents the predicted label of the training sample corresponding to the training network;

[0050] When the loss value meets the preset loss value, the training discriminant network is used as the real discriminant network.

[0051] Optionally, the using the true discriminant network to perform data authenticity assessment on the supplementary data includes:

[0052] Analyzing the data distribution of the precision clustered data and the supplementary data using the true discriminant network to obtain a precision clustered data distribution and a supplementary data distribution;

[0053] Based on the precision clustering data distribution and the supplementary data distribution, the data authenticity of the supplementary data is calculated using the following formula:

[0054] α=∫ A P r (T)R(1,P(U=1|T))da+∫ A P g (T)R(0,1-P(U=1|T))da

[0055] Among them, α represents the authenticity of the data, T represents a specified data in the supplementary data, and P r (T) indicates that T is the precision clustering data distribution, P g (T) indicates that T is the supplementary data distribution, R indicates the loss function, da indicates the differential mapping of T, P(U=1|T) indicates the posterior probability that the specified data T is the precision clustering data, R(1,P(U=1|T) indicates the network loss when the specified data T is the precision clustering data, and R(0,1-P(U=1|T) indicates the network loss when the specified data T is the supplementary data;

[0056] Based on the data authenticity, a data authenticity assessment is performed on the supplementary data.

[0057] Optionally, the using the assessment result of the health assessment and the user information to perform health association analysis on the medical care user includes:

[0058] Determining the associated independent variables and associated dependent variables of the health analysis of the medical and nursing user based on the evaluation result of the health evaluation and the user information;

[0059] Quantitatively encoding the associated independent variable and the associated dependent variable to obtain a coded analysis variable;

[0060] After performing time-related health event analysis and linear health analysis on the medical care user using the coded analysis variables, the health association analysis of the medical care user is completed.

[0061] In a second aspect, the present invention provides a big data health analysis system in a medical and nursing integrated environment, the system comprising:

[0062] A word vector processing module is used to obtain the data to be analyzed in the medical and nursing integration environment, extract the cross-modal features of the data to be analyzed, perform cross-modal fusion on the data to be analyzed based on the cross-modal features to obtain fused data, construct a word vector space for the fused data, and use the word vector space to perform word vectorization processing on the fused data to obtain word vector data;

[0063] A data clustering module, used to construct a small sample of the word vector data, calculate the local density extreme value of the small sample, cluster the word vector data based on the local density extreme value to obtain clustered data, calculate the clustering accuracy of the clustered data, and optimize the clustering of the clustered data based on the clustering accuracy to obtain precision clustered data;

[0064] A data screening module is used to perform data mining on the precision clustering data to obtain supplementary data, construct a true discriminant network for the supplementary data, perform data authenticity evaluation on the supplementary data using the true discriminant network, perform data screening on the supplementary data using the evaluation result of the data authenticity evaluation to obtain screened data, and use the precision clustering data and the screened data as target analysis data;

[0065] The health analysis module is used to query the medical and nursing users corresponding to the target analysis data, extract the user health characteristics of the medical and nursing users in the target analysis data, query the user information of the medical and nursing users, use the user information and the health characteristics to build a user portrait of the medical and nursing users, use the user portrait to perform a health assessment on the medical and nursing users, use the assessment results of the health assessment and the user information to perform a health association analysis on the medical and nursing users, and use the analysis results of the health association analysis to build a health analysis report for the medical and nursing users.

[0066] Compared with the problems described in the background technology, the present invention first obtains the data to be analyzed in the medical and nursing integration environment, and can collect various types of original information related to the medical and nursing integration. This information can come from a variety of channels, such as medical records, elderly care service feedback, elderly health monitoring data, etc., which is helpful to discover potential rules and provide a basis for optimizing the medical and nursing integration model and improving service quality. Furthermore, the embodiment of the present invention performs cross-modal fusion on the data to be analyzed based on the cross-modal features to obtain fused data. The previously extracted cross-modal features are fused according to certain rules and algorithms, and the data from different modalities are organically integrated to generate fused data, so that the data originally scattered in each modality can be presented in a unified manner under a new data representation form, and the fused data contains the key information of each modality, which is more comprehensive. Furthermore, the present invention can simplify the analysis process to a certain extent by constructing a small sample of the word vector data, so that subsequent calculations and processing can be carried out within an acceptable time and resource range, and if the sampling method is reasonable, the small sample can still reflect the overall language of the word vector data. The present invention obtains supplementary data by performing data mining on the precision clustering data, which can help to more comprehensively understand the internal connection between various links of medical and nursing integration, and provide more valuable data support for the subsequent formulation of medical and nursing service plans that are more in line with actual needs and the improvement of the medical and nursing service system, and uses the assessment results of the health assessment and the user information to perform health association analysis on the medical and nursing users, which can deeply explore the internal connection and mutual influence relationship between the health assessment results and the various factors in the user information, such as analyzing whether there is a correlation between factors such as age, gender, and living habits and specific health problems, and mining which factors have a more significant impact on health status through statistical analysis methods, and finding out the health impact mechanism hidden behind the data, thereby improving the level of big data health analysis in the medical and nursing integration environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A flowchart of a big data health analysis method in a medical and nursing integrated environment provided by an embodiment of the present invention;

[0068] Figure 2 A module diagram of a big data health analysis system in a medical and nursing integration environment provided by an embodiment of the present invention.

[0069] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0070] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0071] The embodiment of the present application provides a big data health analysis method in a medical and nursing integration environment. The execution subject of the big data health analysis method for the medical and nursing integration environment includes but is not limited to at least one of the electronic devices such as the server and the terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the big data health analysis method for the medical and nursing integration environment can be executed by software or hardware installed on the terminal device or the server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0072] Embodiment 1:

[0073] Reference Figure 1 FIG. 1 is a flow chart of a big data health analysis method for a medical and nursing integrated environment provided by an embodiment of the present invention. In this embodiment, the big data health analysis method for a medical and nursing integrated environment includes:

[0074] S1. Obtain the data to be analyzed in the medical and nursing integration environment, extract the cross-modal features of the data to be analyzed, perform cross-modal fusion on the data to be analyzed based on the cross-modal features to obtain fused data, construct a word vector space for the fused data, and use the word vector space to perform word vectorization on the fused data to obtain word vector data.

[0075] The embodiments of the present invention can collect various types of original information related to medical-nursing integration by obtaining the data to be analyzed in the medical-nursing integration environment. This information can come from various channels, such as medical records, elderly care service feedback, and elderly health monitoring data, which is helpful to discover potential rules and provide a basis for optimizing the medical-nursing integration model and improving service quality.

[0076] Among them, the medical-nursing integration refers to a new elderly care model that organically integrates medical resources with elderly care resources. The data to be analyzed refers to various information materials covering multiple aspects such as elderly health, elderly care services, medical services, and institutional operation and management, which can reflect the actual situation of medical-nursing integration and help to deeply understand and optimize this model. They can be obtained through the data repository of the medical-nursing integration institution.

[0077] The embodiment of the present invention can find key features that can reflect commonality, association and unique information in different modal data by extracting cross-modal features of the data to be analyzed, and integrate and refine the data of different modalities at the feature level.

[0078] Among them, the cross-modal features refer to the key features that can be extracted across different modalities (such as text, images, audio, video, numerical data, etc.) in a context involving multiple different modal data, reflecting the commonalities, correlations and unique information of each modal data. For example, in a medical and nursing integration environment, there may be text reports on elderly health assessments (describing the elderly's physical condition, disease conditions, etc.), medical images of physical examinations (such as X-rays, CT images, etc. that present images of the internal structure of the body), and real-time monitoring of physiological indicators (numerical data such as blood pressure and blood sugar) These different modal data may exist at the same time.

[0079] As an embodiment of the present invention, the extracting of cross-modal features of the data to be analyzed includes: performing data cleaning on the data to be analyzed to obtain cleaned data, dividing the cleaned data into visual data, text data and numerical data, performing visual feature processing on the visual data to obtain visual features, performing text feature processing on the text data to obtain text features, performing statistical feature processing on the numerical data to obtain numerical features, and determining the cross-modal features of the data to be analyzed based on the visual features, the text features and the numerical features.

[0080] Among them, the visual data refers to information presented in a visual form such as images and videos, the text data refers to content in the form of textual expressions, and the numerical data refers to data composed of specific numbers, often reflecting quantitative situations such as quantities or indicators.

[0081] Optionally, the cleaned data can be obtained by processing the collected data to be analyzed by using conventional data cleaning techniques such as data denoising, removing duplicate values, and filling missing values, and eliminating erroneous, redundant and incomplete parts. The division of the cleaned data into visual data, text data and numerical data can be achieved by classifying the cleaned data into several major types, such as visual (such as images, videos, etc.), text (such as documents, description sentences, etc.) and numerical (such as various indicator values, etc.) according to the data's own form and characteristics, by identifying data formats, content attributes, etc. The visual features can be obtained by extracting features of color, texture, shape, etc. in visual data (such as images) using image feature extraction technology (such as convolutional neural network CNN, etc.). The text features can be obtained by converting text data into vector form or extracting key word frequency, weight and other information using natural language processing technology (such as word vector representation, TF-IDF, etc.). The numerical features can be obtained by analyzing the numerical data with the help of statistical methods (such as calculating statistics such as mean, variance, extreme value, etc.) to obtain numerical features that can reflect its distribution, degree of discreteness, etc.

[0082] The embodiment of the present invention performs cross-modal fusion on the data to be analyzed based on the cross-modal features to obtain fused data. The previously extracted cross-modal features are fused according to certain rules and algorithms, and the data from different modalities are organically integrated to generate fused data, so that the data originally scattered in various modalities can be uniformly presented in a new data representation form. The fused data contains key information of each modality and is more comprehensive.

[0083] Optionally, the process of performing cross-modal fusion on the data to be analyzed based on the cross-modal features to obtain fused data is called: based on the extracted cross-modal features, using appropriate algorithms (such as multimodal fusion algorithms based on deep learning, feature splicing and weighting methods, etc.) to integrate and process the data to be analyzed of different modalities according to multimodal fusion network rules, thereby generating fused data containing multimodal comprehensive information.

[0084] Furthermore, the embodiments of the present invention can quantify text information and mine implicit semantic relationships in the text by constructing the word vector space of the fused data. For example, different ways of expressing similar health problems or service needs can be discovered, thereby better understanding and summarizing various phenomena and problems in the integration of medical care and nursing, and creating conditions for further data mining and knowledge discovery.

[0085] The word vector space refers to a mathematical space concept used to represent the semantic information of words in the field of natural language processing.

[0086] Optionally, the word vector space can be constructed by a Skip-Gram model.

[0087] The embodiment of the present invention utilizes the word vector space to perform word vectorization processing on the fusion data, and the word vector data obtained can be based on these vectors to perform operations such as text classification (distinguishing different types of medical and nursing service feedback), clustering (classifying elderly people with similar health conditions into one category), and association analysis (finding the relationship between health indicators and service needs), etc., to help relevant personnel extract valuable knowledge from massive data, and then continuously improve various practical work of medical and nursing integration, and improve service levels and management efficiency.

[0088] As an embodiment of the present invention, the word vectorization processing of the fused data is performed using the word vector space to obtain word vector data, including: performing word segmentation processing on the fused data to obtain word segmentation data, constructing a vocabulary table of the word segmentation data after word segmentation numbering, using the vocabulary table to construct a word segmentation vector of the word segmentation data after word segmentation numbering, and obtaining the word vector data after inputting the word segmentation vector into the word vector space.

[0089] Optionally, the word segmentation data can use a suitable word segmentation tool (such as Chinese jieba word segmentation, English NLTK, etc.) to split the text content in the fusion data into independent words according to language rules, and obtain the word segmentation data. After the word segmentation data is numbered, the process of constructing the data vocabulary of the word segmentation data is: assigning a unique number to each word segmentation (which can be implemented using binary code), and then counting all different word segmentations to remove duplicates, and summarizing to form a vocabulary containing word segmentations and their numbers. The word segmentation vector can be determined by using the vocabulary as a reference, the vector dimension (corresponding to the number of word segments in the vocabulary), and according to the number and frequency of occurrence of the word segmentation in the vocabulary, the word segmentation vector is constructed by assigning values ​​on the corresponding dimension. The assignment logic is implemented by common programming techniques. The constructed word segmentation vector is implemented according to the established mapping rules and calculation methods of the word vector space, and the input and operation of the vector is implemented by programming, and the corresponding word vector data is finally output.

[0090] S2. Construct a small sample of the word vector data, calculate the local density extreme value of the small sample, cluster the word vector data based on the local density extreme value to obtain clustered data, calculate the clustering accuracy of the clustered data, and optimize the clustering of the clustered data based on the clustering accuracy to obtain accurate clustered data.

[0091] The embodiment of the present invention can simplify the analysis process to a certain extent by constructing a small sample of the word vector data, so that subsequent calculations and processing can be carried out within an acceptable time and resource range. If the sampling method is reasonable, the small sample can still reflect the overall semantic categories, distribution patterns and other characteristics of the word vector data, providing a basis for further mining the value of the data.

[0092] The small sample refers to a relatively small subset of data selected from large-scale overall data according to a certain sampling method (such as random sampling, stratified sampling, systematic sampling, etc.).

[0093] As an embodiment of the present invention, the constructing a small sample of the word vector data includes: randomly sampling the word vector data to obtain sampled word vector data, constructing an initial word vector sample of the word vector data,

[0094] The sample error of the initial word vector sample is calculated using the following formula:

[0095]

[0096] Among them, ME represents the sample error, A represents the number of sampled word vector data, B represents the number of initial word vector samples, and G (i,j) represents the probability that the i-th sampled word vector data is assigned to the j-th initial word vector sample,

[0097] According to the sample error, the initial word vector sample is updated to obtain a small sample.

[0098] Optionally, the sampled word vector data can be obtained by randomly sampling from all the word vector data according to a set sampling ratio (such as 1 to 20, which can be set in combination with actual applications) using programming tools such as random number generators. The initial word vector sample can be obtained by simply arranging and combining the sampled word vector data as an initial word vector sample set, which can be implemented using the List tool in the Python language. The initial word vector sample is updated according to the sample error to obtain a small sample. A reasonable error threshold is set, and the calculated sample error is compared with the threshold size relationship. If the error exceeds the threshold, the initial word vector sample is adjusted and optimized by methods such as adjusting the sampling method, increasing or decreasing the number of samples, etc., until the sample error meets the requirements, and finally a small sample is realized. The error threshold needs to be comprehensively determined in combination with the specific data scale, application scenario, and requirements for sample representativeness. For example, if the word vector data comes from a large-scale text corpus covering the theme of medical and nursing integration, and is subsequently mainly used to preliminarily explore the approximate semantic classification of the text, then the error threshold can be set in a relatively loose range, such as between 0.1 and 0.2.

[0099] The embodiment of the present invention can analyze the data distribution around each data point in the small sample by calculating the local density extreme value of the small sample, and find those extreme value points with relatively high or low local density. These extreme value points are usually located in key positions such as the center or boundary of the data cluster, representing key features of different semantic groups or different distribution areas.

[0100] As an embodiment of the present invention, the calculation of the local density extreme value of the small sample includes: querying the original word vector corresponding to the word vector data corresponding to the small sample and the small sample word vector corresponding to the small sample, constructing the spatial coordinates of the small sample, and calculating the vector distance between the original word vector and the small sample word vector based on the spatial coordinates using the following formula:

[0101]

[0102] X=(x2、y2、z2)

[0103] Y=(x1, y1, z1)

[0104] Among them, VD represents the vector distance, X represents the small sample vector, Y represents the original word vector, (x2, y2, z2) represents a point position of the small sample vector in the spatial coordinates, and (x1, y1, z1) represents a point position of the original word vector in the spatial coordinates.

[0105] The local density extreme value of the small sample is determined according to the vector distance.

[0106] Optionally, the spatial coordinates can determine the coordinate dimensions according to the number of dimensions of the word vector, and correspond the values ​​of each dimension of each word vector to the corresponding spatial coordinate axis respectively. The coordinate assignment is implemented by programming, such as using the array operation of the numpy library in Python to organize the word vector data into the representation of the corresponding spatial coordinates, so as to construct a coordinate system for the small sample in the corresponding space. The local density extreme value of the small sample can be determined according to the vector distance by setting an appropriate distance threshold and a density calculation method (such as taking a certain word vector as the center, and counting the number of other word vectors within the distance threshold as its local density). All calculated vector distance data are traversed by programming, and the local density of each position is counted according to the set density calculation method, and the extreme points with the relatively highest or lowest local density are found out, so as to determine the local density extreme value of the small sample. The distance threshold can be set to 0.5, or it can be set according to the actual application scenario.

[0107] The embodiment of the present invention clusters the word vector data based on the local density extreme value to obtain clustered data that can effectively organize and classify the word vector data, revealing the intrinsic semantic associations and category differences between the words represented by the word vectors. For example, after clustering, you can clearly see which words corresponding to the word vectors belong to the same subject category (such as all related to medical services in medical and nursing integration), which is convenient for subsequent text analysis, topic mining, information retrieval and other applications, and helps to sort out valuable semantic structures from messy word vector data and better understand the information contained in the text data.

[0108] Optionally, the word vector data is clustered based on the local density extreme value, and the clustered data can be obtained by first setting the word vector corresponding to the local density extreme value point as the cluster center, and then calculating the distance between the remaining word vectors and each center, and dividing the cluster members according to the distance.

[0109] Furthermore, the embodiment of the present invention can intuitively understand whether the clustering results obtained based on the previous local density extreme value and clustering operation are reasonable and whether the intrinsic semantic structure of the word vector data is truly excavated by calculating the clustering accuracy of the clustering data. If the clustering accuracy is low, it means that there may be problems with the current clustering method and further adjustment and optimization are required, which provides a quantitative reference standard for subsequent improvements to the clustering data and ensures that the clustering results can effectively serve subsequent text processing tasks.

[0110] As an embodiment of the present invention, the calculating the clustering accuracy of the clustered data includes: calculating the clustering accuracy of the clustered data using the following formula:

[0111]

[0112] Among them, NMI represents the clustering accuracy, T represents the number of clusters of clustered data, n represents the number of clusters generated by the clustering algorithm, and E i Indicates the total number of data points in the i-th cluster in the clustering algorithm result, E j Represents the total number of data points in the jth cluster in the clustered data, Represents the total number of data points that appear in both the i-th cluster of the algorithm result and the j-th class of the true result.

[0113] Furthermore, the embodiment of the present invention optimizes the clustering data based on the clustering accuracy to obtain accurate clustering data, which can continuously improve the quality of clustering, so that the final accurate clustering data can more accurately reflect the true semantic distribution of word vector data.

[0114] Optionally, based on the clustering accuracy, the clustering data is clustered and optimized to obtain accurate clustering data. The clustering accuracy can be evaluated by the Rand Index. If the clustering accuracy is not as expected, the clustering constraints can be increased. For example, in the medical and nursing integration text, in addition to the conventional semantic distance of the word vector, the degree of association between the concepts involved in the word vector in the medical and nursing knowledge graph is additionally considered, and the word vectors that are closely associated in the knowledge graph are preferentially grouped into the same cluster, and then the clustering accuracy is recalculated, or the idea of ​​hierarchical clustering is used for optimization, and the current clustering result is regarded as the initial level, and some clusters are further subdivided or merged. For example, if the analysis finds that the clusters where some word vectors about "rehabilitation training for the elderly" and "daily care for the elderly" are not clearly distinguished, try to split or merge them according to a finer semantic level, and then calculate the new clustering accuracy, and continue to adjust and optimize until the clustering accuracy meets the requirements, thereby obtaining accurate clustering data.

[0115] S3. Perform data mining on the precision clustering data to obtain supplementary data, construct a true discriminant network for the supplementary data, use the true discriminant network to perform data authenticity evaluation on the supplementary data, use the evaluation result of the data authenticity evaluation to perform data screening on the supplementary data to obtain screened data, and use the precision clustering data and the screened data as target analysis data.

[0116] The embodiment of the present invention performs data mining on the precision clustering data to obtain supplementary data, which can help to more comprehensively understand the intrinsic connections between various links of medical and nursing integration, and provide more valuable data support for the subsequent formulation of medical and nursing service plans that are more in line with actual needs and the improvement of the medical and nursing service system.

[0117] Optionally, the data mining of the precision clustering data to obtain supplementary data can be performed through an association rule mining algorithm such as an Apriori algorithm or an FP-Growth algorithm. Taking the Apriori algorithm as an example, the precision clustering data must first be organized into a transaction format suitable for algorithm processing (for example, in a medical and nursing integration scenario, the health data of each elderly person and the corresponding medical and nursing service usage are recorded as a transaction record), and then appropriate support (indicating the frequency of occurrence of a certain rule in a data set) and confidence (indicating the probability of a result occurring when the prerequisite is met) thresholds are set based on business needs and data characteristics. For example, in the integration of medical care and nursing, if you want to explore the association between diseases and rehabilitation programs, you can set the support to 30% (which means that the association rule must appear in at least 30% of the samples) and the confidence to 70% (when suffering from a certain disease, the probability of choosing a specific rehabilitation program must reach 70%). Through programming (such as using Python language and using the mlxtend library to implement the Apriori algorithm function), the algorithm can process the precision clustering data and mine the association rules that meet the threshold requirements. These rules are part of the supplementary data, such as the association information such as "elderly people with hypertension are more likely to choose traditional Chinese medicine physiotherapy rehabilitation programs."

[0118] The embodiment of the present invention can prevent the problem of inaccurate data analysis caused by the fact that manual judgment of the authenticity of the supplementary data is easily affected by professional knowledge limitations and subjective factors by constructing the authenticity discrimination network of the supplementary data.

[0119] Among them, the truth discrimination network is a model or network structure designed to judge the authenticity of data. It belongs to the category of machine learning or deep learning models. Its core function is to output corresponding judgment results based on the input data features through learned discrimination rules and patterns to determine whether the data conforms to the true semantics, logic and data distribution laws, etc., and then distinguish between real data and false data.

[0120] As an embodiment of the present invention, the construction of the true discriminant network of the supplementary data includes: designing a discriminant network architecture for the supplementary data, annotating the supplementary data to obtain annotated data, dividing the annotated data into training data, verification data, and test data, performing preliminary training on the discriminant network architecture using the training data, the verification data, and the test data to obtain a trained discriminant network, and calculating the loss value of the trained discriminant network using the following formula:

[0121]

[0122] Among them, MSE represents the loss value, M represents the number of training data, and H i represents the true label of the training sample corresponding to the training network, Represents the predicted label of the training sample corresponding to the training network;

[0123] When the loss value meets the preset loss value, the training discriminant network is used as the real discriminant network.

[0124] Optionally, the discriminant network architecture for designing the supplementary data may select a suitable network architecture according to the characteristics of the supplementary data. For example, if the supplementary data is an image type (such as medical imaging-related supplementary data in a medical and nursing integration scenario), a convolutional neural network (CNN) may be considered, and the number of convolutional layers, pooling layers, and fully connected layers may be determined; if it is text data (such as medical and nursing service evaluation texts, etc.), a recurrent neural network (RNN) and its variants (such as LSTM, GRU) or a Transformer-based architecture may be selected, and the basic structural parameters such as the order of each layer and the approximate number of neurons may be set. For example, a network including 2 LSTM layers, A network architecture with 128 hidden units in each layer and 2 fully connected layers is used for text data discrimination. Then, a suitable activation function is selected for each layer of neurons to increase the nonlinear ability of the network. For example, ReLU is often used in the hidden layer, and Softmax is used in the output layer for probability output (multi-classification case). At the same time, it is determined whether to add regularization methods such as Dropout to prevent overfitting. For example, the Dropout ratio is set to 0.2, and some neuron connections are randomly discarded to improve the generalization ability of the network. The labeled data can be labeled by the authenticity of the supplementary data based on real and reliable standards and actual experience, such as marking it as "real" or "false". Or use numerical values ​​(such as 1 for true and 0 for false) to distinguish different categories, and clarify the corresponding authenticity label for each supplementary data; the said dividing the labeled data into training data and test data can divide the labeled data according to the ratio of 7:2:1; the said using the training data, the verification data and the test data to preliminarily train the discriminant network architecture to obtain the trained discriminant network can select a suitable optimization algorithm (such as stochastic gradient descent SGD and its variants Adagrad, Adadelta, etc., or the commonly used Adam optimization algorithm), set the initial learning rate (such as 0.001), the training The training rounds (such as 100 rounds) and batch size (such as 32 data points are input for training each time) and other parameters are used. The training data is input into the discriminant network architecture for forward propagation to calculate the prediction results. Then, according to the difference between the prediction results and the true labels (measured by the loss function, such as the cross entropy loss function for classification tasks), the weight parameters of the network are updated by back propagation. This process is iterated repeatedly for training. When the loss value meets the preset loss value, the training discriminant network is used as the real discriminant network by setting a loss threshold and comparing and judging. For example, according to actual needs and previous experimental conditions, a reasonable loss value threshold (such as 0.3, etc., the specific values ​​vary depending on the data and tasks). After calculating the loss value of the training discriminant network each time, compare it with the preset loss value. If the loss value is less than or equal to the preset loss value, it means that the prediction accuracy of the network has reached the expected requirements, and the current training discriminant network is determined as the real discriminant network for subsequent real evaluation of supplementary data; if the loss value is greater than the preset loss value, continue to adjust the network parameters, retrain, calculate the loss value again and make a judgment until the requirements are met. .

[0125] Furthermore, the embodiments of the present invention can quickly and on a large scale screen the supplementary data by utilizing the true discriminant network to perform data authenticity evaluation on the supplementary data, and promptly discover data that seem valuable but do not actually conform to the actual situation of medical-nursing integration, thereby avoiding these data from being mixed into subsequent analysis processes, and ensuring that the data used for further research and decision-making are consistent with the actual medical-nursing scenario, thereby maintaining the scientificity and effectiveness of the entire medical-nursing integration data analysis.

[0126] As an embodiment of the present invention, the use of the true discriminant network to perform data authenticity assessment on the supplementary data includes: using the true discriminant network to analyze the data distribution of the precision clustered data and the supplementary data to obtain the precision clustered data distribution and the supplementary data distribution, and based on the precision clustered data distribution and the supplementary data distribution, using the following formula to calculate the data authenticity of the supplementary data:

[0127] α=∫ A P r (T)R(1,P(U=1|T))da+∫ A P g (T)R(0,1-P(U=1|T))da

[0128] Among them, α represents the authenticity of the data, T represents a specified data in the supplementary data, and P r (T) indicates that T is the precision clustering data distribution, P g (T) indicates that T is the supplementary data distribution, R indicates the loss function, da indicates the differential mapping of T, P(U=1|T) indicates the posterior probability that the specified data T is the precision clustering data, R(1,P(U=1|T) indicates the network loss when the specified data T is the precision clustering data, and R(0,1-P(U=1|T) indicates the network loss when the specified data T is the supplementary data;

[0129] Based on the data authenticity, a data authenticity assessment is performed on the supplementary data.

[0130] Optionally, the use of the true discriminant network to analyze the data distribution of the precision clustered data and the supplementary data to obtain the precision clustered data distribution and the supplementary data distribution can be achieved by using probability density estimation technology (such as kernel density estimation, etc.), inputting the precision clustered data and the supplementary data into the true discriminant network respectively, and allowing the network to output the probability distribution corresponding to each data based on the learned feature representation and data pattern, so as to obtain the precision clustered data distribution and the supplementary data distribution. The data authenticity assessment of the supplementary data based on the data authenticity can be performed by setting a reasonable data authenticity threshold (such as 0.5, etc.), traversing the data authenticity values ​​of the supplementary data through programming, comparing the threshold values, and judging the data with authenticity higher than the threshold as true data, and judging the data with authenticity lower than the threshold as non-true data, thereby completing the authenticity assessment of the supplementary data.

[0131] The embodiment of the present invention screens the supplementary data using the evaluation results of the real evaluation of the data, and the screened data can further purify the supplementary data and improve the overall quality of the data, so that the data finally involved in the analysis are all authentic, reliable and of practical value, providing a strong guarantee for accurately grasping the current situation and development trend of medical and nursing integration.

[0132] Optionally, the evaluation results of the data authenticity evaluation are used to screen the supplementary data, and the screened data can be manually determined according to the specific business scenario, data characteristics and requirements for data authenticity. A suitable data authenticity threshold (such as 0.7, which means that data with authenticity greater than or equal to this value will be retained) is used as the screening standard, and then programming (for example, using Python language and a loop structure) is used to simultaneously traverse the evaluation results (such as authenticity values ​​or true / false category labels) corresponding to each supplementary data obtained by the data authenticity evaluation and the corresponding supplementary data itself, so as to facilitate subsequent screening operations based on the evaluation results. During the traversal process, judgment is made according to the set screening threshold. If the screening is based on the authenticity value, when the authenticity of the supplementary data is greater than or equal to the threshold, the supplementary data is retained; if the screening is based on the category label screening (such as the "real" label), the supplementary data with the corresponding qualified labels are selected, and the data that does not meet the requirements is eliminated. What is finally retained is the screened data.

[0133] The embodiment of the present invention integrates the previously strictly processed precision clustering data and the screening data by using the precision clustering data and the screening data as target analysis data to form a target analysis data set with richer content and more reliable quality. For example, when studying the optimization problem of the medical and nursing integration model, the existing classification information about different types of medical and nursing services in the clustering data, combined with the new associations between medical and nursing services that meet the actual situation and the needs of the elderly mined in the screening data, together constitute a comprehensive and high-quality analysis basis.

[0134] S4. Query the medical and nursing user corresponding to the target analysis data, extract the user health characteristics of the medical and nursing user in the target analysis data, query the user information of the medical and nursing user, use the user information and the health characteristics to construct a user portrait of the medical and nursing user, use the user portrait to perform a health assessment on the medical and nursing user, use the assessment result of the health assessment and the user information to perform a health association analysis on the medical and nursing user, and use the analysis result of the health association analysis to construct a health analysis report of the medical and nursing user.

[0135] The embodiment of the present invention can clarify the specific medical and nursing user individuals or groups associated with the data by querying the medical and nursing users corresponding to the target analysis data, locate information related to specific medical and nursing service objects from massive data, and filter out corresponding data records in a database or data set containing a large amount of medical and nursing data through the user's unique identifier (such as ID number, medical and nursing service account number, etc.).

[0136] The embodiment of the present invention can mine various characteristic information related to the health of medical and nursing users from the target analysis data by extracting the user health characteristics of the medical and nursing users from the target analysis data, such as the elderly's underlying diseases (such as whether they suffer from hypertension, diabetes, etc.), physical function indicators (such as heart rate, blood pressure values, limb mobility, etc.), recent health change trends (such as whether the condition is stable, whether the physical function is improved or declined, etc.) and past medical treatment and rehabilitation records. By integrating and sorting out these fragmented data, the user's health base can be intuitively grasped to formulate targeted medical treatment plans.

[0137] Among them, the user health characteristics refer to various relevant factors and characteristics that can reflect the health status of medical and nursing users, such as basic disease conditions, physical skill indicators, and physiological self-care ability.

[0138] Optionally, the user health characteristics of the medical and nursing users in the target analysis data can be extracted by first determining the fields related to the health characteristics (such as disease name, physical indicator values, etc.), and then extracting the values ​​of the corresponding fields in the acquired user data through programming (dictionary value acquisition, list indexing and other operations can be used in Python), and organizing them into health characteristic data in the form of disease lists, indicator value dictionaries, etc., and then using natural language processing (if it contains text descriptions): If the data contains text descriptions of health conditions, use a word segmentation tool (such as jieba word segmentation) to first segment the words, and then extract key health words based on a keyword extraction algorithm (such as TF-IDF), and add them to the health feature set to achieve implementation.

[0139] The embodiment of the present invention can collect other relevant information of the medical and nursing user in addition to health characteristics by querying the user information of the medical and nursing user, including personal basic information (such as age, gender, residential address, etc.), family situation (such as contact information of family members, whether there are children to take care of, etc.), economic situation (such as whether medical insurance is covered, ability to pay out of pocket, etc.) and living habits (such as dietary preferences, daily activity patterns, smoking and drinking conditions, etc.), etc., to further enrich the comprehensive understanding of the medical and nursing user.

[0140] Optionally, the user information of medical and nursing users can be queried through database association query (relational database) using SQL JOIN operation to integrate multi-table data.

[0141] The embodiment of the present invention constructs a user portrait of the medical and nursing user by utilizing the user information and the health characteristics of the health characteristics user. The extracted health characteristics and the queried user information can be integrated, summarized and visualized (in the form of labels, charts, etc.) to form a user portrait that can comprehensively and intuitively reflect the characteristics of the medical and nursing user.

[0142] The user portrait is a tool for providing a comprehensive and concrete description of a specific user group or individual.

[0143] Optionally, the user portrait can be obtained by merging health features and user information, dividing them into different categories (such as basic information, health status, living habits, etc.), and using a programming language (Python can be obtained through dictionary nesting, etc.) to construct a structured data object.

[0144] The embodiment of the present invention utilizes the user portrait to conduct a health assessment of the medical and nursing user. Based on the health characteristics, living habits, underlying diseases and other information presented in the user portrait, professional medical knowledge and health assessment models (such as health risk assessment models, elderly living ability assessment scales, etc.) are used to conduct a comprehensive analysis and evaluation of the current health status of the medical and nursing user, determine the level of their health level (such as good, average, poor, etc.), identify potential health risks (such as the risk of cardiovascular disease onset, the risk of falls, etc.) and the degree of impact on daily living ability.

[0145] Optionally, the use of the user portrait to conduct a health assessment of the medical and nursing user can be achieved by selecting daily living activities (ADL) scales, health risk prediction models, etc. based on the age, major health problems, etc. of the medical and nursing user. These scales can be obtained from a professional assessment tool library or constructed based on statistics and machine learning methods. Then, the data such as relevant health characteristics and living habits in the portrait are formatted according to the model requirements and substituted into the model (by programming to call the model calculation function, etc.) to obtain health assessment results such as health level and disease risk probability.

[0146] The embodiment of the present invention uses the assessment results of the health assessment and the user information to perform health association analysis on the medical and nursing users, which can deeply explore the intrinsic connection and mutual influence relationship between the health assessment results and the factors in the user information. For example, it analyzes whether there is a correlation between factors such as age, gender, and living habits and specific health problems (such as the incidence of a certain chronic disease, rehabilitation effects, etc.), and uses statistical analysis methods (such as correlation analysis, regression analysis, etc.) to find out which factors have a more significant impact on health status and find out the health impact mechanism hidden behind the data.

[0147] As an embodiment of the present invention, the use of the assessment results of the health assessment and the user information to perform a health association analysis on the medical and nursing user includes: determining the associated independent variables and the associated dependent variables of the health analysis of the medical and nursing user based on the assessment results of the health assessment and the user information, quantizing and encoding the associated independent variables and the associated dependent variables to obtain coded analysis variables, and using the coded analysis variables to perform time-related health event analysis and linear health analysis on the medical and nursing user to complete the health association analysis of the medical and nursing user.

[0148] Among them, the associated independent variables refer to the various factors or variables that are considered to have an impact on the health status of the users, which are selected from the user information of the medical and nursing users in the context of health association analysis of the medical and nursing users. For example, in the association analysis of the risk of cardiovascular disease among medical and nursing users, age, smoking, weekly exercise duration, etc. can all be used as associated independent variables. The associated dependent variables refer to specific indicators that reflect the health status of medical and nursing users. They come from the evaluation results of health assessments and are target variables that users hope to explain, predict or explore their relationship with other factors through analysis. For example, whether the medical and nursing user suffers from a certain chronic disease (expressed as "yes" or "no", which belongs to the classification) The time-related health event analysis refers to the process of focusing on specific events that occur when the health status of medical and nursing users changes over time, and studying the influence of various factors (i.e., associated independent variables) on these health-related time events. The linear health analysis refers to the process of constructing and analyzing this linear relationship model by using statistical analysis methods such as linear regression, assuming that there is a linear relationship between the health status of medical and nursing users (represented by associated dependent variables) and various influencing factors (associated independent variables).

[0149] Optionally, the determination of the associated independent variables and associated dependent variables of the health analysis of the medical and nursing user based on the assessment results of the health assessment and the user information can be achieved by carefully combing through the key indicators that can reflect the health status in the health assessment results (such as disease risk level, physical function score, etc.) as the associated dependent variables, which represent the health results to be analyzed and explained, and screening out factors that may affect health (such as age, living habits, economic status, etc.) from the user information as associated independent variables; the quantitative encoding of the associated independent variables and the associated dependent variables to obtain the coded analysis variables can be achieved by assigning category values ​​to categorical independent variables (such as gender, smoking status, etc.) using digital coding methods (such as 0 and 1); assigning values ​​to ordered categorical variables in sequence; using the coded analysis variables to perform time-related health event analysis on the medical and nursing user; and using the coded analysis variables to perform The analysis of time-related health events can be performed using survival analysis methods (such as the Kaplan-Meier method and the Cox proportional hazard regression model). In the corresponding statistical software or programming environment (such as the R language, Python combined with related libraries), the coded analysis variables are input as required for calculation to analyze the impact of each variable on the time of health events. The linear health analysis of the medical and nursing users using the coded analysis variables can be performed by using linear regression analysis, forming a feature matrix of the independent variables in the coded analysis variables, and using the dependent variable as the target vector. After dividing the data set, the model training and prediction are performed using related classes such as the scikit-learn library of Python. By analyzing the positive and negative, size and evaluation indicators (such as mean square error, etc.) of the regression model coefficients, the linear influence relationship of the independent variable on the dependent variable (health status related indicators) is determined to complete the linear health analysis, and then the overall health association analysis is completed.

[0150] The embodiment of the present invention constructs the health analysis report of the medical and nursing user by utilizing the analysis results of the health association analysis, and can present the detailed results of the relationship between the health status of the medical and nursing user and various factors obtained by the health association analysis in a standardized and clear report form. The content covers basic information of the user, health assessment conclusions, health risk prompts, key factors affecting health and related situations, targeted health suggestions, etc., to form a comprehensive and instructive health analysis document.

[0151] Optionally, the health analysis report can be completed by first developing a report template that includes sections such as basic information, health assessment conclusions, health risk warnings, correlation analysis, and health recommendations, and then filling the corresponding contents obtained in the previous steps into the corresponding sections in turn, and then using a document generation library (such as Python's docx library) to organize the report contents into a Word format document.

[0152] Embodiment 2:

[0153] like Figure 2 As shown, it is a system functional module diagram of a big data health analysis method for a medical and nursing integrated environment provided by an embodiment of the present invention.

[0154] The big data health analysis system 200 in the medical and nursing integration environment of the present invention can be installed in an electronic device. According to the functions implemented, the big data health analysis system in the medical and nursing integration environment may include a word vector processing module 201, a data clustering module 202, a data screening module 203 and a health analysis module 204. The module of the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0155] In the embodiment of the present invention, the functions of each module / unit are as follows:

[0156] The word vector processing module 201 is used to obtain the data to be analyzed in the medical and nursing integration environment, extract the cross-modal features of the data to be analyzed, perform cross-modal fusion on the data to be analyzed based on the cross-modal features to obtain fused data, construct a word vector space for the fused data, and use the word vector space to perform word vectorization processing on the fused data to obtain word vector data;

[0157] The data clustering module 202 is used to construct a small sample of the word vector data, calculate the local density extreme value of the small sample, cluster the word vector data based on the local density extreme value to obtain clustered data, calculate the clustering accuracy of the clustered data, and optimize the clustering of the clustered data based on the clustering accuracy to obtain precision clustered data;

[0158] The data screening module 203 is used to perform data mining on the precision clustering data to obtain supplementary data, construct a true discriminant network for the supplementary data, use the true discriminant network to perform data authenticity evaluation on the supplementary data, use the evaluation result of the data authenticity evaluation to perform data screening on the supplementary data to obtain screened data, and use the precision clustering data and the screened data as target analysis data;

[0159] The health analysis module 204 is used to query the medical and nursing users corresponding to the target analysis data, extract the user health characteristics of the medical and nursing users in the target analysis data, query the user information of the medical and nursing users, use the user information and the health characteristics to build a user portrait of the medical and nursing users, use the user portrait to perform a health assessment on the medical and nursing users, use the assessment results of the health assessment and the user information to perform a health association analysis on the medical and nursing users, and use the analysis results of the health association analysis to build a health analysis report for the medical and nursing users.

[0160] In detail, each module in the big data health analysis system 200 in the medical and nursing integration environment of the embodiment of the present invention is used in the same manner as above. Figure 1 The same technical means are used in the big data health analysis method in the medical and nursing integration environment described in the text, and can produce the same technical effects, so I will not go into details here.

[0161] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0162] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0163] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0164] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0165] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0166] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A big data health analysis method in a medical and nursing integrated environment, characterized by: The method comprises: Acquire the data to be analyzed in the medical and nursing integration environment, extract the cross-modal features of the data to be analyzed, perform cross-modal fusion on the data to be analyzed based on the cross-modal features to obtain fused data, construct a word vector space of the fused data, and use the word vector space to perform word vectorization processing on the fused data to obtain word vector data; Constructing a small sample of the word vector data, calculating a local density extreme value of the small sample, clustering the word vector data based on the local density extreme value to obtain clustered data, calculating the clustering accuracy of the clustered data, and performing clustering optimization on the clustered data based on the clustering accuracy to obtain precision clustered data; Performing data mining on the precision clustering data to obtain supplementary data, constructing a true discriminant network for the supplementary data, performing data authenticity evaluation on the supplementary data using the true discriminant network, performing data screening on the supplementary data using an evaluation result of the data authenticity evaluation to obtain screened data, and using the precision clustering data and the screened data as target analysis data; Query the medical and nursing users corresponding to the target analysis data, extract the user health characteristics of the medical and nursing users in the target analysis data, query the user information of the medical and nursing users, use the user information and the health characteristics to build a user portrait of the medical and nursing users, use the user portrait to perform a health assessment on the medical and nursing users, use the assessment result of the health assessment and the user information to perform a health association analysis on the medical and nursing users, and use the analysis result of the health association analysis to build a health analysis report for the medical and nursing users.

2. The big data health analysis method for medical and nursing integration environment according to claim 1, characterized in that: The extracting cross-modal features of the data to be analyzed includes: Performing data cleaning on the data to be analyzed to obtain cleaned data; dividing the cleaned data into visual data, text data and numerical data; Performing visual feature processing on the visual data to obtain visual features; Performing text feature processing on the text data to obtain text features; Performing statistical feature processing on the numerical data to obtain numerical features; Determine the cross-modal features of the data to be analyzed according to the visual features, the text features, and the numerical features.

3. The big data health analysis method for medical and nursing integration environment according to claim 1 is characterized in that: The using the word vector space to perform word vectorization processing on the fused data to obtain word vector data includes: Performing word segmentation processing on the fused data to obtain word segmentation data; After the word segmentation data are numbered, a vocabulary table of the word segmentation data is constructed; Using the vocabulary, constructing a segmentation vector of the segmentation data after segmentation numbering; After the word segmentation vector is input into the word vector space, word vector data is obtained.

4. The big data health analysis method for medical and nursing integration environment according to claim 1 is characterized in that: The step of constructing a small sample of the word vector data includes: Randomly sampling the word vector data to obtain sampled word vector data; Constructing an initial word vector sample of the word vector data; The sample error of the initial word vector sample is calculated using the following formula: Among them, ME represents the sample error, A represents the number of sampled word vector data, B represents the number of initial word vector samples, and G (i,j) represents the probability that the i-th sampled word vector data is assigned to the j-th initial word vector sample, According to the sample error, the initial word vector sample is updated to obtain a small sample.

5. The big data health analysis method for medical and nursing integration environment according to claim 1 is characterized in that: The calculating of the local density extreme value of the small sample comprises: Query the original word vector corresponding to the word vector data corresponding to the small sample and the small sample word vector corresponding to the small sample; Constructing the spatial coordinates of the small sample; The vector distance between the original word vector and the small sample word vector is calculated based on the spatial coordinates using the following formula: X=(x2、y2、z2) Y=(x1, y1, z1) Where VD represents the vector distance, X represents the small sample vector, Y represents the original word vector, (x2, y2, z2) represents a point position of the small sample vector in the spatial coordinates, and (x1, y1, z1) represents a point position of the original word vector in the spatial coordinates; The local density extreme value of the small sample is determined according to the vector distance.

6. The big data health analysis method for medical and nursing integration environment according to claim 1, characterized in that: The calculating the clustering accuracy of the clustered data includes: The clustering accuracy of the clustered data is calculated using the following formula: Among them, NMI represents the clustering accuracy, T represents the number of clusters of clustered data, n represents the number of clusters generated by the clustering algorithm, and E i Indicates the total number of data points in the i-th cluster in the clustering algorithm result, E j Represents the total number of data points in the jth cluster in the clustered data, Represents the total number of data points that appear in both the i-th cluster of the algorithm result and the j-th class of the true result.

7. The big data health analysis method for medical and nursing integration environment according to claim 1 is characterized in that: The constructing of a true discriminant network for the supplementary data comprises: Designing a discriminative network architecture for the supplementary data; Performing data labeling on the supplementary data to obtain labeled data; Dividing the labeled data into training data, verification data, and test data; Using the training data, the verification data and the test data, the discriminant network architecture is preliminarily trained to obtain a trained discriminant network; The loss value of the training discriminant network is calculated using the following formula: Among them, MSE represents the loss value, M represents the number of training data, and H i represents the true label of the training sample corresponding to the training network, Represents the predicted label of the training sample corresponding to the training network; When the loss value meets the preset loss value, the training discriminant network is used as the real discriminant network.

8. The big data health analysis method for medical and nursing integration environment according to claim 1, characterized in that: The using the true discriminant network to perform data authenticity evaluation on the supplementary data includes: Analyzing the data distribution of the precision clustered data and the supplementary data using the true discriminant network to obtain a precision clustered data distribution and a supplementary data distribution; Based on the precision clustering data distribution and the supplementary data distribution, the data authenticity of the supplementary data is calculated using the following formula: Among them, α represents the authenticity of the data, T represents a specified data in the supplementary data, and P r (T) indicates that T is the precision clustering data distribution, P g (T) indicates that T is the supplementary data distribution, R indicates the loss function, da indicates the differential mapping of T, P(U=1|T) indicates the posterior probability that the specified data T is the precision clustering data, R(1,P(U=1|T) indicates the network loss when the specified data T is the precision clustering data, and R(0,1-P(U=1|T) indicates the network loss when the specified data T is the supplementary data; Based on the data authenticity, a data authenticity assessment is performed on the supplementary data.

9. The big data health analysis method for medical and nursing integration environment according to claim 1, characterized in that: The step of using the evaluation result of the health evaluation and the user information to perform health association analysis on the medical and nursing user includes: Determining the associated independent variables and associated dependent variables of the health analysis of the medical and nursing user based on the evaluation result of the health evaluation and the user information; Quantitatively encoding the associated independent variable and the associated dependent variable to obtain a coded analysis variable; After performing time-related health event analysis and linear health analysis on the medical care user using the coded analysis variables, the health association analysis of the medical care user is completed.

10. A big data health analysis system applied in the medical and nursing integration environment, characterized in that: The system comprises: A word vector processing module is used to obtain the data to be analyzed in the medical and nursing integration environment, extract the cross-modal features of the data to be analyzed, perform cross-modal fusion on the data to be analyzed based on the cross-modal features to obtain fused data, construct a word vector space for the fused data, and use the word vector space to perform word vectorization processing on the fused data to obtain word vector data; A data clustering module, used to construct a small sample of the word vector data, calculate the local density extreme value of the small sample, cluster the word vector data based on the local density extreme value to obtain clustered data, calculate the clustering accuracy of the clustered data, and optimize the clustering of the clustered data based on the clustering accuracy to obtain precision clustered data; A data screening module is used to perform data mining on the precision clustering data to obtain supplementary data, construct a true discriminant network for the supplementary data, perform data authenticity evaluation on the supplementary data using the true discriminant network, perform data screening on the supplementary data using the evaluation result of the data authenticity evaluation to obtain screened data, and use the precision clustering data and the screened data as target analysis data; The health analysis module is used to query the medical and nursing users corresponding to the target analysis data, extract the user health characteristics of the medical and nursing users in the target analysis data, query the user information of the medical and nursing users, use the user information and the health characteristics to build a user portrait of the medical and nursing users, use the user portrait to perform a health assessment on the medical and nursing users, use the assessment results of the health assessment and the user information to perform a health association analysis on the medical and nursing users, and use the analysis results of the health association analysis to build a health analysis report for the medical and nursing users.

Images